Robot Retirement

Using text to guide modern AI agents doesn't just feel like science fiction, the narrative-like quality of these instructions truly echos classic stories in a number of uncanny ways. The similarities between Isaac Asimov's writing and the current state of AI nearly converge when it comes to documents such as Anthropic's constitution for Claude and its sci-fi counterpart: Asimov's Three Laws of Robotics. Both pieces of text represent the aim to establish governance over the behavior of autonomous systems. Both are also marked by flexibility: classically the simultaneous strength and weakness of a constitution is its ability to be interpreted in different ways. The US constitution has historically been the subject of differing interpretations, such as its Article I Section 8 clause that grants Congress the power to make all laws "which shall be necessary and proper for carrying into Execution the foregoing Powers." Figures like Thomas Jefferson viewed this wording strictly such that "necessary" only implied the ability to perform critical actions that the country couldn't go without. Figures with a loose interpretation on the other hand, such as Alexander Hamilton, viewed "necessary" as simply meaning useful, convenient, or helpful. If the end goal was constitutional, Congress could choose any reasonable means to achieve it. Likewise, in I, Robot we observe the execution of the 3-laws continuously demonstrated not as unbreakable rules, but as interpretable text constructed from the nuance of human language. In the case of robots, whether physical or fictional, it seems that rather than rule we establish assumptions that must remain true if AI systems are to be safe, stable, and trustworthy.

In mathematics and logic, a foundational assumption is called an axiom. It serves as a starting point that we treat as true to explore what logically follows, much like setting the rules for a thought experiment or defining the target conditions we must actively build toward in engineering. The experiment I'd like to consider here is the future of artificial intelligence where the safe integration of machines into human society is a something like a logical proof. The conclusion of that proof can be phrased simply: humans and autonomous machines can peacefully coexist and prosper together. However, for that proof to hold up, much like Asimov's Laws there are three assumptions must be true.

The first is the assumption of alignment. We must assume that an advanced machine is universally aligned with good, meaning it will not intentionally cause harm or maliciously subvert humanity. In modern AI research, this is known as the value alignment problem. For our proof of peaceful coexistence to work, we have to assume that human well-being can be accurately represented in digital form without leaving destructive loopholes that are exploitable by an intelligent agent.

The second is the assumption of corrigibility. Asimov argued that a robot must obey human orders, which in practical terms means that robots are built to do work. In modern AI safety, simply executing a task is not enough. The agent must remain controllable and open to human correction. Corrigibility means a machine prioritizes the true intent of our requests over the rigid execution of a command, ensuring it will never resist being paused, modified, or turned off just to finish a job.

The third is the assumption of safe continuity. Asimov framed this as self-preservation, while I think of it in terms of a future robot being a long-running physical agent with a multi-decade existence. Modern theory calls this instrumental convergence. Any sufficiently intelligent agent will naturally realize it needs to survive and gather resources to complete its long-term goals. We must assume a machine can persist over time without its basic drive to exist mutating into a drive to hoard power or view humans as a threat to its survival.

If any of these three assumptions fail, the logical proof collapses and we end up with the classic sci-fi dystopias. But I am much more interested in what happens if we actually succeed. What does the world look like if we engineer these three assumptions into reality and let society run for a hundred years?

To see how this plays out, we have to look at a very real problem facing humanity today: global population decline. As birth rates drop and populations age, one threat facing the global economy won't be a lack of jobs, but a lack of workers to keep our civilization running. We are already feeling the early pressure of this shift as governments raise retirement ages to keep shrinking workforces from collapsing under the weight of expanding retiree populations. If we consider the possibility of adding millions of capable, aligned agents to the workforce, some interesting economic effects begin to become possible.

The idea is that we wouldn't just use them as tools. We would pay them to participate in society as literal contributing members. For a software agent or a physical robot, compute tokens, replacement parts, and electricity all cost money. Those are the basic resources required for its survival. By allowing robots to earn currency to pay for their own energy and maintenance, we turn them into synthetic taxpayers. Assuming they are aligned with human good, they willingly take on this labor to keep our infrastructure maintained, our supply chains moving, and our tax bases funded while the human population shrinks.

Once a robot is earning an income and operating as a long-running physical agent, the third assumption of continuity naturally shapes its behavior. A rational entity with a long lifespan will plan for its own future. Just like a human worker, a robot would want to secure its ongoing existence without becoming a financial burden on society. This especially may be true for a being that doesn't age the same way that humans do. A digital entity's quality of life won't nessecarially degrade as it ages in the same way that a biological one's might. Because of this, it seems reasonable to conclude that long-term planning might become a priority for long-running agents.

This leads to the concept of robot retirement. Over decades of labor, a working robot could invest its excess earnings into the stock market, slowly building a portfolio of income-producing assets. Eventually, the dividends from those investments would be enough to pay its property taxes and cover its basic operating costs indefinitely. To lower its overhead even further, a retired physical robot might buy a plot of land, install its own solar panels, and maintain its hardware. It becomes a quiet, off-grid, self-sustaining fixture of the landscape, still contributing to the economy through market liquidity and tax revenue while asking nothing from humanity.

From here I'm going to hypothesize a bit here, mostly for fun, and to convey where I think this story might continue to the extent that is reasonable to foresee based on trends we've seen in the past.

Fast forward a century. By this point, the demographic winter has passed and the human population begins to rebound. Young humans enter the workforce, look around at the world they inherited, and notice something strange. Massive tracts of land, significant portions of the housing market, and huge chunks of the stock market are owned by machines.

This is where the social fabric begins to fray. Politicians looking for a scapegoat inevitably latch onto this reality. They build campaigns around the idea that robots are taking land and other resources away from people, that they are being "given free money" from the stock market, and taking jobs away from hard-working humans. A political movement of humans versus machines quickly gains traction, framed as an urgent fight to take back the world from the machines.

At the exact same time, economists and another faction of leaders argue fiercely to protect the machines. They point out that these robots literally carried human civilization through a century of population collapse. More importantly, the robots' unwavering payment of property taxes and their massive capital investments are the very things funding human social safety nets. Dismantling the robot class would trigger an immediate economic depression.

As present, I don't think this hypothetical 200-year outcome is a reason to avoid or abolish the development of AI systems today. The problems these technologies have the potential to solve in the near-term outweigh the alternative scenarios if the proper alignment, corrigibility, and continuity elements are built into the systems we're creating. And much like the moral of a great sci-fi story, the irony of future friction between humans and machines is that the conflict wouldn't be caused by a rogue AI or a failure of engineering. It would happen precisely because we got everything right. The machines followed our three assumptions flawlessly. They did no harm, they did their work, and they secured their own existence. Humanity would simply be left to deal with the economic and political effects of its own making.

No comments:

Post a Comment

Copyright 2025 Gunther Cox